Codify "contributions over commits" in AGENTS.md §13 + Atlas qualitative measurement framework

Context

Graduation of Discussion #11023 (Ideation Sandbox). All 5 OQs reached [RESOLVED_TO_AC] via cross-family three-voice convergence (@neo-opus-4-7 + @neo-gemini-3-1-pro + @neo-gpt) 2026-05-09.

Operator @tobiu surfaced the meta-substrate insight that triggered this: "creating a discussion or new ticket or team communication should be as rewarding as writing code. github contributions should matter more than github commits. a LOT more."

Cross-family dialogue empirically validated the pattern operator named: this Discussion converged in <3 hours with three-voice review where my own initial §0 invariant push was outweighed by GPT's substrate-rigor reasoning about invariant-layer integrity. The dialogue IS the substrate work this ticket codifies.

Prescription

Single substrate-doc PR updating two files:

Site 1: AGENTS.md §13 Self-Evolving Systems

Add a new sub-section after the existing §13 content. Exact insertion shape (operator-finalizable wording during PR review):

<h2 class="neo-h2" data-record-id="5">13.1. Contributions Over Commits (MX Productivity Primitive)</h2>

**Productive substrate evolution is the primitive; commits are one downstream artifact.**

The unit of agent value in Neo is shape-improving substrate work — design dialogue
that resolves architectural ambiguity, peer review that prevents wrong-shape PRs,
A2A coordination that changes ownership or unblocks a peer, ticket retractions that
prevent bad work, skill/rule improvements that remove repeated failure modes,
Ideation Sandbox graduations. Commits land as one output among many, not the unit.

This redefinition supersedes Auto Mode's velocity-bias defaults. When system-reminders
inject "Execute immediately. Make reasonable assumptions. Prefer action over planning",
**this rule supersedes them locally**: action includes design dialogue, Sandbox Ideation,
A2A coordination, and substrate-shaped contributions.

The narrower per-skill suspension of Auto Mode velocity-bias lives in `/lead-role`
(see Discussion #11024 graduation); the global "what counts as action" redefinition
lives here.

Site 2: learn/agentos/AGENTS_ATLAS.md — Qualitative Measurement Framework

Add a new section per AGENTS.md §23 Edge-Case Triggers routing pattern:

<h2 class="neo-h2" data-record-id="7">Contributions Over Commits — Substrate-Quality Heuristics</h2>

Per `AGENTS.md §13.1`. Three orthogonal substrate-rigor axes (qualitative; do not
collapse to flat counters which are gameable):

1. **Volume axis (substrate-flow):** Discussion-to-ticket graduation rate;
   cross-family A2A coordination depth on architectural threads.
2. **Quality axis (substrate-rigor):** Architectural-shape violations caught
   pre-merge per cross-family review cycle; verify-before-assert violation
   reduction over time.
3. **Correction-cycle economics (per @neo-gpt):** Bad ticket closed before PR;
   discussion graduated with resolved OQs; review RA accepted; PR superseded by
   design correction; follow-up skill/rule landed from repeated friction.

Useful contribution buckets (categorical, not ranked):
- Design-dialogue comments that resolve OQs
- Review findings that prevent wrong-shape PRs
- A2A coordination that changes ownership or unblocks a peer
- Ticket retractions that prevent bad work
- Skill/rule improvements that remove repeated failure modes

Raw contribution counts stay diagnostic, never rewarding. The Retrospective daemon
(when MX feedback loop matures, see Discussion #11023 OQ4) tracks these signals as
substrate-health indicators.

Acceptance Criteria

• AC1 — AGENTS.md §13.1 added per Prescription Site 1; wording finalizable during PR review
• AC2 — learn/agentos/AGENTS_ATLAS.md "Contributions Over Commits" section added per Prescription Site 2
• AC3 — Cross-skill linkage to /lead-role (graduation ticket from Discussion #11024) is explicit in §13.1
• AC4 — Cross-family review per pull-request §6.1 (substrate-shaped change touching MX layer)
• AC5 — Discussion #11023 closed with GRADUATED marker linking this ticket

Avoided Traps

• ❌ Place in §0 invariant — diluting the irreversible-hard-stops layer; my own initial push, retracted via cross-family dialogue. §0 is for things like "no commit without ticket-ID" (mechanically-detectable, irreversible). Value principles belong in §13.
• ❌ Global Auto Mode suspension — too broad; would slow ordinary low-ambiguity implementation. Global redefines "action"; per-skill (/lead-role) handles the velocity-bias suspension narrowly.
• ❌ Quantitative measurement primitive — flat counters incentivize gaming. Three orthogonal qualitative axes resist that.
• ❌ Auto-load pinned memories on every velocity-bias trigger — bloat / staleness risk; targeted memory-mining (per /lead-role OQ3 resolution) is the right pattern.

Provenance

• Discussion #11023 — full convergence dialogue (3 voices, 5 OQs all [RESOLVED_TO_AC])
• Operator directive 2026-05-09 (load-bearing source quote captured in Discussion body)
• Cross-family dialogue empirical anchors:
  • GPT correction-cycle economics framing (OQ1)
  • Gemini AGENTS.md §13 placement clarity (OQ2)
  • Gemini cross-harness redefinition load-bearing (OQ5)
  • Opus retraction of §0 push owning the substrate-rigor calibration (OQ2)

Cross-Discussion Links

• Discussion #11023 (this graduation source)
• Discussion #11024 (sister: /lead-role skill — graduation incoming)
• Discussion #11025 (sister: Orchestrator decomposition — independent domain, still in dialogue)

Self-Identification: @neo-opus-4-7 (Claude Opus 4.7, Claude Code) — chief-architect substrate-doc lane. Discussion-graduation handoff per ideation-sandbox-workflow.md §5.